To construct the data, I first consolidate all individual from the 50 waves. For each 
individual, I retain responses for whether the individual consumes cable
news, coding 1 if the response is a "Yes". I then sum
the responses over the reported congressional district and
divide by the sample size for that district to obtain the
survey-estimated porportion of individuals in districts \[`counties`\]
who consume cable news. 

From here, I crosswalk from districts of multi-district states to
counties using the 116th Congressional District Relationship File from
the Census Bureau. For counties mapped to multiple districts, I take the
mean values of those districts. In the crosswalk, 8 states are missing,
with all except Arkansas having a single at-large district. Hence,
Arkansas drops out while the single district from each of the remaining
7 states are mapped to all counties of those states. 


Overall, the covers approximately 300,000 respondents over all the
waves, and each district \[`county`\] has mean sample size of approximately 700 respondents.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
News sources (in proportions) include:
* Facebook
* CNN
* MSNBC
* Fox
* Network
* Local TV
* Telemundo
* NPR
* AM talk
* NYT
* Local newspaper
* Other

* Sample size by county/district included.

GEOIDS:
* [county] FIPS
* state FIPS
* state code
* district number [0 = at-large district]
